[PATCH v2 2/2] arm64/module: Use text-poke API for late relocations.

[Song Liu]

On Tue, Apr 22, 2025 at 5:27 PM Dylan Hatch <dylanbhatch at google.com> wrote:
>
> On Mon, Apr 21, 2025 at 11:25 PM Song Liu <song at kernel.org> wrote:
> >
> > On Mon, Apr 21, 2025 at 5:35 PM Song Liu <song at kernel.org> wrote:
> > >
> > > On Fri, Apr 11, 2025 at 6:10 PM Dylan Hatch <dylanbhatch at google.com> wrote:
> > > >
> > > > To enable late module patching, livepatch modules need to be able to
> > > > apply some of their relocations well after being loaded. In this
> > > > scenario, use the text-poking API to allow this, even with
> > > > STRICT_MODULE_RWX.
> > > >
> > > > This patch is largely based off commit 88fc078a7a8f6 ("x86/module: Use
> > > > text_poke() for late relocations").
> > > >
> > > > Signed-off-by: Dylan Hatch <dylanbhatch at google.com>
> >
> > Could you please share how you test this?
> >
>
> For context, we enable livepatch for arm64 by porting this RFC series
> (along with other internal patches) into our kernel:
> https://lore.kernel.org/all/20230202074036.507249-1-madvenka@linux.microsoft.com/.
>
> The way I tested this patch is: with STRICT_MODULE_RWX, load a module
> and a livepatch that touches that module (in either order), and
> confirm the kernel doesn't crash.
>
> Without this patch, a crash is caused in apply_relocate_add() if both
> a module and a livepatch that touches the module are both loaded. This
> happens through one of two code paths:
>
>   1. If the module is already loaded when the livepatch is applied,
> through the module_init() callback.
>   2. If the module is loaded after the livepatch is applied, through
> prepare_coming_module().
>
> In both scenarios, the livepatch module's text is already RX-only.

Thanks for sharing the information!

Could you please help test this set with [1]? This is a different approach
than the one by Madhavan. We were hoping to ship it soon.

Thanks,
Song

[1] https://lore.kernel.org/live-patching/20250320171559.3423224-1-song@kernel.org/